ARROW-1990: [JS] C++ Refactor, Add DataFrame

This PR moves the `Table` class out of the Vector hierarchy and adds optimized dataframe operations to it. Currently implements an optimized `scan()` method, `filter(predicate)`, `count()`, and `countBy(column_name)` (only works on dictionary-encoded columns). Some usage examples, based on the file generated by `js/test/data/tables/generate.py`: ``` js > let table = Table.from(...); > table.count() 1000000 > table.filter(col('lat').gteq(0)).count() 499718 > table.countBy('origin').toJSON() { Charlottesville: 166839, 'New York': 166251, 'San Francisco': 166642, Seattle: 166659, 'Terre Haute': 166756, 'Washington, DC': 166853 } > table.filter(col('lng').gteq(0)).countBy('origin').toJSON() { Charlottesville: 83109, 'New York': 83221, 'San Francisco': 83515, Seattle: 83362, 'Terre Haute': 83314, 'Washington, DC': 83479 } ``` There are performance tests for the dataframe operations, to run them you must first generate the test data by running `npm run create:perfdata`. The PR also includes @trxcllnt's refactor of the JS implementation to make it more closely resemble the C++ implementation. This refactor resolves multiple JIRAs: ARROW-1903, ARROW-1898, ARROW-1502, ARROW-1952 (partially), and ARROW-1985 Author: Paul Taylor <paul.e.taylor@me.com> Author: Brian Hulette <brian.hulette@ccri.com> Author: Brian Hulette <hulettbh@gmail.com> Closes #1482 from TheNeuralBit/table-scan-perf and squashes the following commits: 52f1e0e [Brian Hulette] <, > are not commutative, misc cleanup 04b1838 [Brian Hulette] even more table tests 16b9ccb [Brian Hulette] Merge pull request #4 from trxcllnt/js-cpp-refactor fe300df [Paul Taylor] fix closure es5/umd toString() iterator 3d5240a [Paul Taylor] fix more externs 10c48ad [Paul Taylor] Merge branch 'table-scan-perf' of github.com:ccri/arrow into js-cpp-refactor dbe7f81 [Brian Hulette] Add more Table unit tests 1910962 [Brian Hulette] Add optional bind callback to scan 5bdf17f [Brian Hulette] Fix perf 8cf2473 [Brian Hulette] Merge remote-tracking branch 'origin/master' into table-scan-perf 4a41b18 [Paul Taylor] add src/predicate to the list of exports we should save from uglify 5a91fab [Paul Taylor] add more view, predicate externs f6adfb3 [Brian Hulette] Create predicate namespace f7bb0ed [Paul Taylor] Merge branch 'table-scan-perf' of github.com:ccri/arrow into js-cpp-refactor e148ee4 [Paul Taylor] Merge branch 'extern-woes' into js-cpp-refactor 25cdc4a [Paul Taylor] add src/predicate to the list of exports we should save from uglify dc7c728 [Paul Taylor] add more view, predicate externs 25e6af7 [Brian Hulette] Create predicate namespace 579ab1f [Brian Hulette] Merge pull request #2 from trxcllnt/js-cpp-refactor f3cde1a [Paul Taylor] fix lint 9769773 [Paul Taylor] fix vector perf tests 016ba78 [Brian Hulette] Merge pull request #1 from trxcllnt/js-cpp-refactor 272d293 [Paul Taylor] Merge pull request #4 from ccri/empty-table 7bc7363 [Brian Hulette] Fix exception for empty Table 8ddce0a [Paul Taylor] check bounds in getChildAt(i) to avoid NPEs f1dead0 [Paul Taylor] compute chunked nested childData list correctly 18807c6 [Paul Taylor] rename ChunkData's fields so it's more clear they're not semantically similar to other similarly named fields 7e43b78 [Paul Taylor] add test:integration npm script a5f200f [Paul Taylor] Merge pull request #3 from ccri/table-from-struct c8cd286 [Brian Hulette] Add Table.fromStruct a00415e [Brian Hulette] Fix perf 54d4f5b [Paul Taylor] lazily allocate table and recordbatch columns, support NestedView's getChildAt(i) method in ChunkedView 40b3638 [Paul Taylor] run integration tests with local data for coverage stats fe31ee0 [Paul Taylor] slice the flat data values before returning an iterator of them e537789 [Paul Taylor] make it easier to run all integration tests from local data c0fd2f9 [Paul Taylor] use the dictionary of the last chunked vector list for chunked dictionary vectors e33c068 [Paul Taylor] Merge pull request #2 from ccri/fixed-size-list 5bb63af [Brian Hulette] Don't read OFFSET vector for FixedSizeList 614b688 [Paul Taylor] add asEpochMs to date and timestamp vectors 87334a5 [Paul Taylor] Merge branch 'table-scan-perf' of github.com:ccri/arrow into js-cpp-refactor b7f5bfb [Paul Taylor] rename numRows to length, add table.getColumn() e81082f [Paul Taylor] export vector views, allow cloning data as another type 700a47c [Paul Taylor] export visitors e859e13 [Paul Taylor] fix package.json bin entry 0620cfd [Brian Hulette] use Math.fround 0126dc4 [Brian Hulette] Don't recompute total length e761eee [Brian Hulette] Rename asJSON to toJSON 6c91ed4 [Paul Taylor] Merge branch 'master' of github.com:apache/arrow into js-cpp-refactor-merge_with-table-scan-perf d2b18d5 [Paul Taylor] Merge remote-tracking branch 'ccri/table-scan-perf' into js-cpp-refactor-merge_with-table-scan-perf f3f3b86 [Paul Taylor] rename table.ts to recordbatch.ts in preparation for merging latest e3f629d [Paul Taylor] fix rest of the mangling issues fa7c17a [Paul Taylor] passing all tests except es5 umd mangler ones e20decd [Brian Hulette] Add license headers edcbdbe [Brian Hulette] cleanup 20717d5 [Brian Hulette] Fixed countBy(string) 7244887 [Brian Hulette] Add table unit tests... 6719147 [Brian Hulette] Add DataFrame.countBy operation 2f4a349 [Brian Hulette] Minor tweaks 2e118ab [Brian Hulette] linter a788db3 [Brian Hulette] Cleanup a9fff89 [Brian Hulette] Move Table out of the Vector hierarchy 1d60aa1 [Brian Hulette] Moved DataFrame ops to Table. DataFrame is now an interface e8979ba [Brian Hulette] Refactor DataFrame to extend Vector<StructRow> 6a41d68 [Brian Hulette] clean up table benchmarks 2744c63 [Brian Hulette] Remove Chunked/Simple DataFrame distinction aa999f8 [Brian Hulette] Add DictionaryVector optimization for equals predicate 4d9e8c0 [Brian Hulette] Add concept of predicates for filtering dataframes 796f45d [Brian Hulette] add DataFrame filter and count ops 30f0330 [Brian Hulette] Add basic DataFrame impl ... a1edac2 [Brian Hulette] Add perf tests for table scans d18d915 [Paul Taylor] fix struct and map rows 61dc699 [Paul Taylor] WIP -- refactor types to closer match arrow-cpp 62db338 [Paul Taylor] update dependencies and add es6+ umd targets to jest transform ignore patterns to fix ci 6ff18e9 [Paul Taylor] ship es2015 commonJS in main package to avoid confusion 74e828a [Paul Taylor] fix typings issues (ARROW-1903)

Author: trxcllnt

js/bin/integration.js

@@ -17,6 +17,8 @@
 // specific language governing permissions and limitations
 // under the License.
 
+var fs = require('fs');
+var glob = require('glob');
 var path = require('path');
 var gulp = require.resolve(path.join(`..`, `node_modules/gulp/bin/gulp.js`));
 var child_process = require(`child_process`);
@@ -29,12 +31,14 @@ var optionList = [
  {
  type: String,
  name: 'arrow', alias: 'a',
- description: 'The Arrow file to read/write'
+ multiple: true, defaultValue: [],
+ description: 'The Arrow file[s] to read/write'
  },
  {
  type: String,
  name: 'json', alias: 'j',
- description: 'The JSON file to read/write'
+ multiple: true, defaultValue: [],
+ description: 'The JSON file[s] to read/write'
  }
 ];
 
@@ -66,20 +70,60 @@ function print_usage() {
  process.exit(1);
 }
 
-if (!argv.arrow || !argv.json || !argv.mode) {
+let jsonPaths = argv.json;
+let arrowPaths = argv.arrow;
+
+if (!argv.mode) {
+ return print_usage();
+}
+
+let mode = argv.mode.toUpperCase();
+if (mode === 'VALIDATE' && !jsonPaths.length) {
+ jsonPaths = glob.sync(path.resolve(__dirname, `../test/data/json/`, `*.json`));
+ if (!arrowPaths.length) {
+ [jsonPaths, arrowPaths] = jsonPaths.reduce(([jsonPaths, arrowPaths], jsonPath) => {
+ const { name } = path.parse(jsonPath);
+ for (const source of ['cpp', 'java']) {
+ for (const format of ['file', 'stream']) {
+ const arrowPath = path.resolve(__dirname, `../test/data/${source}/${format}/${name}.arrow`);
+ if (fs.existsSync(arrowPath)) {
+ jsonPaths.push(jsonPath);
+ arrowPaths.push(arrowPath);
+ console.log('-j', jsonPath, '-a', arrowPath, '\\');
+ }
+ }
+ }
+ return [jsonPaths, arrowPaths];
+ }, [[], []]);
+ }
+} else if (!jsonPaths.length) {
  return print_usage();
 }
 
-switch (argv.mode.toUpperCase()) {
+switch (mode) {
  case 'VALIDATE':
+ const args = [`test`, `-i`].concat(argv._unknown || []);
+ jsonPaths.forEach((p, i) => {
+ args.push('-j', p, '-a', arrowPaths[i]);
+ });
  child_process.spawnSync(
- gulp,
- [`test`, `-i`].concat(process.argv.slice(2)),
+ gulp, args,
  {
  cwd: path.resolve(__dirname, '..'),
  stdio: ['ignore', 'inherit', 'inherit']
  }
  );
+ // for (let i = -1, n = jsonPaths.length; ++i < n;) {
+ // const jsonPath = jsonPaths[i];
+ // const arrowPath = arrowPaths[i];
+ // child_process.spawnSync(
+ // gulp, args.concat(['-j', jsonPath, '-a', arrowPath]),
+ // {
+ // cwd: path.resolve(__dirname, '..'),
+ // stdio: ['ignore', 'inherit', 'inherit']
+ // }
+ // );
+ // }
  break;
  default:
  print_usage();

js/gulp/argv.js

@@ -15,20 +15,22 @@
 // specific language governing permissions and limitations
 // under the License.
 
+const fs = require('fs');
+const glob = require('glob');
+const path = require('path');
+
 const argv = require(`command-line-args`)([
  { name: `all`, type: Boolean },
  { name: 'update', alias: 'u', type: Boolean },
  { name: 'verbose', alias: 'v', type: Boolean },
  { name: `target`, type: String, defaultValue: `` },
  { name: `module`, type: String, defaultValue: `` },
  { name: `coverage`, type: Boolean, defaultValue: false },
- { name: `json_file`, alias: `j`, type: String, defaultValue: null },
- { name: `arrow_file`, alias: `a`, type: String, defaultValue: null },
  { name: `integration`, alias: `i`, type: Boolean, defaultValue: false },
  { name: `targets`, alias: `t`, type: String, multiple: true, defaultValue: [] },
  { name: `modules`, alias: `m`, type: String, multiple: true, defaultValue: [] },
- { name: `sources`, alias: `s`, type: String, multiple: true, defaultValue: [`cpp`, `java`] },
- { name: `formats`, alias: `f`, type: String, multiple: true, defaultValue: [`file`, `stream`] },
+ { name: `json_files`, alias: `j`, type: String, multiple: true, defaultValue: [] },
+ { name: `arrow_files`, alias: `a`, type: String, multiple: true, defaultValue: [] },
 ], { partial: true });
 
 const { targets, modules } = argv;
@@ -38,4 +40,25 @@ argv.module && !modules.length && modules.push(argv.module);
 (argv.all || !targets.length) && targets.push(`all`);
 (argv.all || !modules.length) && modules.push(`all`);
 
+if (argv.coverage && (!argv.json_files || !argv.json_files.length)) {
+
+ let [jsonPaths, arrowPaths] = glob
+ .sync(path.resolve(__dirname, `../test/data/json/`, `*.json`))
+ .reduce((paths, jsonPath) => {
+ const { name } = path.parse(jsonPath);
+ const [jsonPaths, arrowPaths] = paths;
+ ['cpp', 'java'].forEach((source) => ['file', 'stream'].forEach((format) => {
+ const arrowPath = path.resolve(__dirname, `../test/data/${source}/${format}/${name}.arrow`);
+ if (fs.existsSync(arrowPath)) {
+ jsonPaths.push(jsonPath);
+ arrowPaths.push(arrowPath);
+ }
+ }));
+ return paths;
+ }, [[], []]);
+
+ argv.json_files = jsonPaths;
+ argv.arrow_files = arrowPaths;
+}
+
 module.exports = { argv, targets, modules };

js/gulp/closure-task.js

@@ -36,7 +36,7 @@ const closureTask = ((cache) => memoizeTask(cache, function closure(target, form
  const src = targetDir(target, `cls`);
  const out = targetDir(target, format);
  const entry = path.join(src, mainExport);
- const externs = path.join(src, `${mainExport}.externs`);
+ const externs = path.join(`src/Arrow.externs.js`);
  return observableFromStreams(
  gulp.src([
 /* external libs first --> */ `node_modules/tslib/package.json`,
@@ -46,7 +46,6 @@ const closureTask = ((cache) => memoizeTask(cache, function closure(target, form
  `node_modules/text-encoding-utf-8/package.json`,
  `node_modules/text-encoding-utf-8/src/encoding.js`,
 /* then sources globs --> */ `${src}/**/*.js`,
-/* and exclusions last --> */ `!${src}/Arrow.externs.js`,
  ], { base: `./` }),
  sourcemaps.init(),
  closureCompiler(createClosureArgs(entry, externs)),
@@ -60,14 +59,15 @@ const closureTask = ((cache) => memoizeTask(cache, function closure(target, form
 }))({});
 
 const createClosureArgs = (entry, externs) => ({
+ externs,
  third_party: true,
  warning_level: `QUIET`,
  dependency_mode: `STRICT`,
  rewrite_polyfills: false,
- externs: `${externs}.js`,
  entry_point: `${entry}.js`,
  module_resolution: `NODE`,
- // formatting: `PRETTY_PRINT`, debug: true,
+ // formatting: `PRETTY_PRINT`,
+ // debug: true,
  compilation_level: `ADVANCED`,
  allow_method_call_decomposing: true,
  package_json_entry_names: `module,jsnext:main,main`,

js/gulp/package-task.js

@@ -45,10 +45,11 @@ const createMainPackageJson = (target, format) => (orig) => ({
  ...createTypeScriptPackageJson(target, format)(orig),
  name: npmPkgName,
  main: mainExport,
+ types: `${mainExport}.d.ts`,
  module: `${mainExport}.mjs`,
  dist: `${mainExport}.es5.min.js`,
  [`dist:es2015`]: `${mainExport}.es2015.min.js`,
- [`@std/esm`]: { esm: `mjs` }
+ [`@std/esm`]: { esm: `mjs`, warnings: false, sourceMap: true }
 });
 
 const createTypeScriptPackageJson = (target, format) => (orig) => ({
@@ -63,18 +64,20 @@ const createTypeScriptPackageJson = (target, format) => (orig) => ({
 
 const createScopedPackageJSON = (target, format) => (({ name, ...orig }) =>
  conditionallyAddStandardESMEntry(target, format)(
- packageJSONFields.reduce(
- (xs, key) => ({ ...xs, [key]: xs[key] || orig[key] }),
- { name: `${npmOrgName}/${packageName(target, format)}`,
- version: undefined, main: `${mainExport}.js`, types: `${mainExport}.d.ts`,
- dist: undefined, [`dist:es2015`]: undefined, module: undefined, [`@std/esm`]: undefined }
- )
+ packageJSONFields.reduce(
+ (xs, key) => ({ ...xs, [key]: xs[key] || orig[key] }),
+ {
+ name: `${npmOrgName}/${packageName(target, format)}`,
+ version: undefined, main: `${mainExport}.js`, types: `${mainExport}.d.ts`,
+ dist: undefined, [`dist:es2015`]: undefined, module: undefined, [`@std/esm`]: undefined
+ }
+ )
  )
 );
 
 const conditionallyAddStandardESMEntry = (target, format) => (packageJSON) => (
- format !== `esm`
- ? packageJSON
- : { ...packageJSON, [`@std/esm`]: { esm: `js` } }
+ format !== `esm` && format !== `cls`
+  ?  packageJSON
+  : { ...packageJSON, [`@std/esm`]: { esm: `js`, warnings: false, sourceMap: true } }
 );
 

js/gulp/test-task.js

@@ -44,15 +44,15 @@ const testOptions = {
 const testTask = ((cache, execArgv, testOptions) => memoizeTask(cache, function test(target, format, debug = false) {
  const opts = { ...testOptions };
  const args = !debug ? [...execArgv] : [...debugArgv, ...execArgv];
- args.push(`test/${argv.integration ? `integration/*` : `unit/*`}`);
+ if (!argv.coverage) {
+ args.push(`test/${argv.integration ? `integration/*` : `unit/*`}`);
+ }
  opts.env = { ...opts.env,
  TEST_TARGET: target,
  TEST_MODULE: format,
- JSON_PATH: argv.json_file,
- ARROW_PATH: argv.arrow_file,
  TEST_TS_SOURCE: !!argv.coverage,
- TEST_SOURCES: JSON.stringify(Array.isArray(argv.sources) ? argv.sources : [argv.sources]),
- TEST_FORMATS: JSON.stringify(Array.isArray(argv.formats) ? argv.formats : [argv.formats]),
+ JSON_PATHS: JSON.stringify(Array.isArray(argv.json_files) ? argv.json_files : [argv.json_files]),
+ ARROW_PATHS: JSON.stringify(Array.isArray(argv.arrow_files) ? argv.arrow_files : [argv.arrow_files]),
  };
  return !debug ?
  child_process.spawn(jest, args, opts) :

js/gulp/typescript-task.js

@@ -34,7 +34,7 @@ const typescriptTask = ((cache) => memoizeTask(cache, function typescript(target
  const tsProject = ts.createProject(path.join(`tsconfig`, tsconfigFile), { typescript: require(`typescript`) });
  const { stream: { js, dts } } = observableFromStreams(
  tsProject.src(), sourcemaps.init(),
- tsProject(ts.reporter.fullReporter(true))
+ tsProject(ts.reporter.defaultReporter())
  );
  const writeDTypes = observableFromStreams(dts, gulp.dest(out));
  const writeJS = observableFromStreams(js, sourcemaps.write(), gulp.dest(out));
@@ -52,12 +52,12 @@ function maybeCopyRawJSArrowFormatFiles(target, format) {
  return Observable.empty();
  }
  return Observable.defer(async () => {
- const outFormatDir = path.join(targetDir(target, format), `format`, `fb`);
+ const outFormatDir = path.join(targetDir(target, format), `fb`);
  await del(path.join(outFormatDir, '*.js'));
  await observableFromStreams(
- gulp.src(path.join(`src`, `format`, `fb`, `*_generated.js`)),
+ gulp.src(path.join(`src`, `fb`, `*_generated.js`)),
  gulpRename((p) => { p.basename = p.basename.replace(`_generated`, ``); }),
  gulp.dest(outFormatDir)
  ).toPromise();
  });
-}
+}

js/gulp/uglify-task.js

@@ -29,7 +29,7 @@ const webpack = require(`webpack`);
 const { memoizeTask } = require('./memoize-task');
 const { Observable, ReplaySubject } = require('rxjs');
 const UglifyJSPlugin = require(`uglifyjs-webpack-plugin`);
-const esmRequire = require(`@std/esm`)(module, { cjs: true, esm: `js` });
+const esmRequire = require(`@std/esm`)(module, { cjs: true, esm: `js`, warnings: false });
 
 const uglifyTask = ((cache, commonConfig) => memoizeTask(cache, function uglifyJS(target, format) {
 
@@ -84,11 +84,20 @@ module.exports = uglifyTask;
 module.exports.uglifyTask = uglifyTask;
 
 const reservePublicNames = ((ESKeywords) => function reservePublicNames(target, format) {
- const publicModulePath = `../${targetDir(target, format)}/${mainExport}.js`;
- return [
- ...ESKeywords,
- ...reserveExportedNames(esmRequire(publicModulePath))
+ const src = targetDir(target, format);
+ const publicModulePaths = [
+ `../${src}/data.js`,
+ `../${src}/type.js`,
+ `../${src}/table.js`,
+ `../${src}/vector.js`,
+ `../${src}/util/int.js`,
+ `../${src}/predicate.js`,
+ `../${src}/recordbatch.js`,
+ `../${src}/${mainExport}.js`,
  ];
+ return publicModulePaths.reduce((keywords, publicModulePath) => [
+ ...keywords, ...reserveExportedNames(esmRequire(publicModulePath, { warnings: false }))
+ ], [...ESKeywords]);
 })(ESKeywords);
 
 // Reflect on the Arrow modules to come up with a list of keys to save from Uglify's
@@ -104,8 +113,8 @@ const reserveExportedNames = (entryModule) => (
  .map((name) => [name, entryModule[name]])
  .reduce((reserved, [name, value]) => {
  const fn = function() {};
- const ownKeys = value && Object.getOwnPropertyNames(value) || [];
- const protoKeys = typeof value === `function` && Object.getOwnPropertyNames(value.prototype) || [];
+ const ownKeys = value && typeof value === 'object' && Object.getOwnPropertyNames(value) || [];
+ const protoKeys = typeof value === `function` && Object.getOwnPropertyNames(value.prototype || {}) || [];
  const publicNames = [...ownKeys, ...protoKeys].filter((x) => x !== `default` && x !== `undefined` && !(x in fn));
  return [...reserved, name, ...publicNames];
  }, []

js/gulp/util.js

@@ -87,7 +87,7 @@ const ESKeywords = [
  // EventTarget
  `addListener`, `removeListener`, `addEventListener`, `removeEventListener`,
  // Arrow properties
- `low`, `high`, `data`, `index`, `field`, `validity`, `columns`, `fieldNode`, `subarray`,
+ `low`, `high`, `data`, `index`, `field`, `columns`, 'numCols', 'numRows', `values`, `valueOffsets`, `nullBitmap`, `subarray`
 ];
 
 function taskName(target, format) {
@@ -108,14 +108,13 @@ function targetDir(target, format) {
 
 function logAndDie(e) {
  if (e) {
- console.error(e);
  process.exit(1);
  }
 }
 
 function observableFromStreams(...streams) {
- const pumped = streams.length <= 1 ? streams[0]
-   : pump(...streams, logAndDie);
+ if (streams.length <= 0) { return Observable.empty(); }
+ const pumped = streams.length <= 1 ? streams[0] : pump(...streams, logAndDie);
  const fromEvent = Observable.fromEvent.bind(null, pumped);
  const streamObs = fromEvent(`data`)
  .merge(fromEvent(`error`).flatMap((e) => Observable.throw(e)))